RST Reader - Improved Role Support

[bergey]

These changes make Roles behave more like the RST definition, and I hope more useful. I've also added warnings when the RST reader ignores syntax.

• use Code constructor for :literal: role
• add custom role name to classes (when possible)
• allow role directive without parent - use Span constructor

Commit e784d4b refactors custom role handling, but shouldn't change behavior (except order of classes). I think it makes the code easier to understand, but feedback is welcome.

@merijn, do you have time to take a look? I think you were the last to work on custom roles. In particular, I've changed the type of stateRstCustomRoles. Before my changes, the base role was computed when defining a new custom role, and ignored when rendering the role. I don't see any reason (other than maybe debugging) why we need to store both of these. I also don't see how the parent role could depend on the Attrs. Does this look OK, or am I breaking something?

This differs from rst2xml.py in (at least) the following ways, which I'm willing to fix if they bother people. I suspect they're not issues in practice.

• If a custom role inherits from a custom role, both role names are included in the classes
• The :span: role is recognized as built-in, to implement custom roles with unspecified parent
• For unknown roles, pandoc outputs the quoted text, but not the role markup; rst2xml outputs both

[jgm]

This looks promising to me -- let's wait a few days to see if @merijn chimes in.

[merijn]

At a glance it looks fine to me. To be honest, I'm not really sure why stateRstCustomRoles was implemented like that, it made sense to me at the time, but I'd been messing around for awhile, so I'm pretty sure that's just accidental artefact of previous attempts.

[jgm]

Thanks!